I'm surprised to see no discussion about how well Jan's program is doing. Jan, are you pleased with it? I think I read earlier that you didn't feel it would lose any games, but I think 45-7 is a great record.

Looks like Xander has its number, and Jan beat it also.

It was an honor to lose to it. ;-)

Rex

Jan's Dvonn program clearly plays very strongly. It is stronger than both other Dvonn programs I played so far. It can be beaten, but to do so it is necessary to decide the game early on, in the endgame it cannot be beaten. As Jan states in the info, the only weakness is its capability to recognize stones being cut off. At the moment my scoring against the program is 4-2 (in my advantage), but since the program is continuously improving, winning will become more difficult in the future. (Luckily my play is also slowly improving ;-)

In any case: a congratulation to Jan is certainly well deserved and I cannot wait until the program will be released.

Xander

> Jan's Dvonn program clearly plays very strongly.

Thanks.

> It is stronger than both other Dvonn programs I played so far.

That's cool:)

> It can be beaten, but to do so it is necessary to decide the game early on, in the endgame it cannot be beaten.

Very true, the setup has enormous weaknesses as shows my win over my own program.

> As Jan states in the info, the only weakness is its capability to recognize stones being cut off.

After it has played 100 games it will 'rest' for a while waiting for a major re-training session after which especially the cut-off problem will be much less. Also the speed will be improved a lot so that it is able to look 1 or 2 moves further ahead.

> At the moment my scoring against the program is 4-2 (in my advantage), but since the program is continuously improving, winning will become more difficult in the future. (Luckily my play is also slowly improving ;-)

I hope at some stage the program will be able to play such good that human players start *seeing* new tactics or strategies that the program can only numerize but not understand. This will help increasing the level of play of the top-players.

> In any case: a congratulation to Jan is certainly well deserved and I cannot wait until the program will be released.

It is already released, read the forum! Don't forget to download the .NET framework first!

> I'm surprised to see no discussion about how well Jan's program is doing.

Now you mention it, I'm positively surprised that only hot_sync_karen complained about it being a program.

> Jan, are you pleased with it?

The start is here, although I've far from implemented all the improvements i've thought off by now, no time…I'm reasonably pleased, however winning from the top-players still is problematic.

> I think I read earlier that you didn't feel it would lose any games, but I think 45-7 is a great record.

You're correct. I had a version that beat me personally 100%, I simply had no chance. Unfortunately that training-database had some problems and was destroyed while I tried to resque it. After that I found several annoying bugs of which 1 I still haven't been able to pinpoint/correct.

> Looks like Xander has its number, and Jan beat it also.

Xander is trying (and succeeding) to beat it without using anti-program tactics. This is smart since he'll learn a lot from it himself! I beat my program just to show that it still has serious flaws!

> It was an honor to lose to it. ;-)

I'm happy that so many players play against it without resigning or complaining. It gives good info to be able to improve it.

I don't play Dvonn, so it's not an issue for me. If someone develops an experimental program in, say, Twixt, I wouldn't mind playing it in rating tournaments. Especially if it was clearly labeled as a program and it signed up for the tournament before I did. But I don't think I'd like it if the program was competing for the monthly cups and championships.

does anyone have deeplinks/urls to where i can find dvonn programs?

The links to Jan de Graafs Dvonn program, you can find in this forum under the title 'Dvonn practice program', there you can download and try it.

A second proogram is Dvonner from Matthias Bodenstein (matztias @BSW):

http://matztias.de/spiele/dvonner/dvonner-e.htm

It can also be downloaded.

I know a 3rd program is being made, but that one is not yet available.

When you are playing Jan's program…how deep does it search? Is there a time limit Jan puts on it, or a depth of search?

My program searches at least 6 half moves ahead. If it has reached that depth within 1 minute it fills up that 1 minute. Then it 'guesses' how long it'll take to compute the remaining moves for the depth it is currently computing on and if that 'guess' is less than 5 minutes it goes on (mostly). Sometimes when I'm actively watching I let it compute deeper or longer.

During the setup it reaches 8ply always, up to 16ply at the end of the setup. In the stacking-phase it doesn't reach beyond 6 ply generally for the first 10-12 half moves, then it gets “easy” enough to do more than 6ply, actually quite quickly 8 or more ply. The endgame is perfectly computed with 23 stack (incl. de dvonn stacks) or less. Sometimes in easy positions it starts with 25 or 26 stacks left on the board.